On 03/06/2013 10:10 AM, Konstantin Boudnik wrote:
I want to throw in the following discussion topic.
We are clearly having integration issues when putting together
bleeding edge Hadoop based stacks. I don't want to argue about the
roots of the phenomena not about how it can be fixed. The purpose of
this discussion is different.
There is an ongoing discussion happening at
http://s.apache.org/hadoop-stablization
and http://is.gd/HX5JA2
I think this is the time for community to start doing something and
improving the situation, helping our users in the passing. I am sure
the effort outlined below would beneficial for individuals as well as
commercial vendors alike.
Now, think Ubuntu - arguably the _most_ successful community driven
Linux distribution in the world. There are two trains of the releases
happening all the time: STS and LTS. The former have about 6 months of
support in the form of updates and is used as a technology preview for
the next LTS release. The latter is good for 3 years and normally is a
way more stable then STS releases.
Currently, BigTop is focusing on the latest (and apparently not
greatest) of its component releases. While sometimes it gives certain
guarantees that A of X.y version will work with B of W.v version, it
isn't much of reconciliation to the users and vendors who simply want
to have a stable reference implementation of a Hadoop stack that has
been defined and supported by the community, with clear path for the
updates and testing put out in the open.
Because of the nature of the differences between Hadoop 1 and Hadoop 2
we can even try to run two different LTS for both kinds of stack.
I believe there's not much more to say about it, except that this is,
in my opinion, a good way to establish our project as de-facto go-to
place for community driven Hadoop based stacks and a focal point for
the integration in the ASF storage and analytics projects.
Let's the discussion begin.
Cos
Hi,
What was preventing anyone from offering such stack until now?
I was under the impression this is what the branch 0.3.X is all about.
There was some interest, but apparently not enough to get releases out.
So I am all for it but I am not sure building the consensus for a LTS
stamp will be enough if we don't get enough interest/people/time to have
stable releases out in the first place.
Why not:
* Backport applicable bugfixes from trunk into a stable branch of your
choice (let say 0.3.X)
* Update minor version (supposedly bugfix only) of the downstream
components in the stable branch of your choice (let say 0.3.X)
* Add more tests to bigtop if necessary
* Run the tests
* Start a vote on a release
Rinse and repeat.
Note that I agree with you, having some LTS releases would be great. But
beside doing the work, I fail to see what is preventing us from doing so
right now.
Thanks,
Bruno
